Research on Indication Extraction Method of Continuous Renal Replacement Therapy Based on Deep Learning

In the treatment of patients with continuous renal replacement therapy, doctors need to manually extract relevant indications from a large number of unstructured medical records, which has been faced with the problems of large and scattered information and low efficiency of medical records. To this end, a deep learning framework combining RoBERTa model and BiLSTM-CRF module is constructed. The model first fuses the RoBERTa pre-trained language model with the semantic features generated by the Transformer layer to capture the rich semantic features in the context of the case text, and then these semantic representations are passed to the BiLSTM model. The model further extracts deep-level text features, and finally generates the optimal label sequence through the CRF layer as the prediction result. Finally, the method is experimentally analyzed on the electronic medical record data set of adult patients receiving continuous renal replacement therapy in a provincial people's hospital. The F1 score is as high as 92.9%, and a good recognition effect is achieved, which proves that the model can effectively identify the entities of continuous renal replacement therapy indications.